Catherine M. Curtin

Summary: Peripheral nerve injuries can profoundly impact patients’ functionality and quality of life. Classification systems such as those proposed by Seddon and Sunderland have provided valuable frameworks for understanding the nature and severity of nerve injuries. However, recent emerging evidence has shed light on an underrecognized category of nerve injury. This article provides a review of nerve injury classification systems and presents information on the Sunderland zero injury, a milder category of nerve injury characterized by immediate recovery upon decompression. We present a focused Sunderland zero literature review and a case report of a 26-year-old with a brachial plexus injury who developed secondary proximal median nerve compression. Intraoperative somatosensory-evoked potentials and motor-evoked potentials were obtained immediately before and after surgical decompression to dynamically assess nerve function. Preoperatively, the patient demonstrated median nerve motor deficits despite interval recovery from the brachial plexus injury. Immediately following decompression at the level of the lacertus fibrosus, increased motor-evoked potential amplitude and improved somatosensory-evoked potential waveform quality were demonstrated. Early postoperative examination confirmed functional improvement without the expected temporal profile of axonal regeneration. Sunderland zero represents a reversible ischemic conduction block without demyelination or axonal loss. Recognition of this nerve injury category is critical, as timely intervention may result in immediate clinical improvement and meaningful functional recovery and improved patient outcomes. Further research is needed to refine diagnostic and therapeutic strategies to identify Sunderland zero injuries.

Objectives: To develop and demonstrate a practical post-hoc, model-agnostic fairness auditing tool that integrates group-level and individual-level fairness assessment for clinical prediction models. Materials and Methods: We developed EquiLense, a fairness auditing tool that operationalizes three components: group fairness evaluation using established and novel group-level disparity metrics, individual fairness assessment, and the Mean Predicted Probability Difference (MPPD), a novel metric quantifying prediction inconsistencies between clinically similar patients across demographic groups. We applied EquiLense to EHR data from 59,047 surgical patients across post-surgical delirium and 30-day readmission prediction models, and demonstrated MPPD on two external benchmarks with documented bias: the COMPAS recidivism dataset and the UCI Adult Income dataset. Results: Applied to delirium prediction (AUROC 0.78), MPPD identified prediction inconsistencies between clinically similar patients that varied systematically across demographic groups. Among patients clinically similar to White patients, Asian patients exhibited the largest disparity (MPPD difference = 0.045). In external benchmarks, MPPD assigned the highest disparity scores to African American defendants in COMPAS and to female workers in UCI Adult Income, consistent with documented bias in both datasets. Discussion: Current fairness evaluation approaches address group-level disparities or individual-level consistency in isolation, creating blind spots in algorithmic fairness assessment. EquiLense bridges this gap by integrating established fairness metrics with MPPD in a single post-hoc tool, enabling comprehensive and interpretable fairness auditing without model retraining. Conclusion: EquiLense provides a practical approach to fairness auditing in clinical AI by combining group and individual fairness perspective through MPPD, a novel metric demonstrated on both clinical and benchmark datasets. Lay Summary: Clinical prediction models are increasingly used across clinical care, research, and decision-making, but they can produce unfair outcomes for patients. Fairness can be evaluated at the group level, by comparing outcomes across demographic categories, or at the individual level, by assessing whether clinically similar patients receive consistent predictions. Most existing tools address one perspective but not both together. We developed EquiLense, a post-hoc fairness auditing tool that bridges these two perspectives. EquiLense operationalizes three components: (1) group fairness evaluation using using multiple group-level disparity metrics; (2) individual fairness assessment measuring prediction consistency among clinically similar patients; and (3) the Mean Predicted Probability Difference (MPPD), a novel metric that quantifies how differently a model treats two patients who are clinically similar but belong to different demographic groups. We applied EquiLense to surgical outcome prediction models using data from over 59,000 patients, and validated MPPD on two external datasets with well-documented bias (COMPAS recidivism and UCI Adult Income). MPPD detected known disparities in both benchmarks. In our clinical application, we found that models can achieve strong overall performance while still producing systematically different predictions for clinically similar patients from different sensitive attributes. EquiLense requires no model retraining, applies to any clinical prediction model, and produces interpretable outputs grounded in clinically familiar features, making it directly applicable in real-world healthcare settings.

Purpose: The purpose of this scoping review is to analyze the application of artificial intelligence (AI) in ultrasound (US) imaging for diagnosing carpal tunnel syndrome (CTS), with an aim to explore the potential of AI in enhancing diagnostic accuracy, efficiency, and patient outcomes by automating tasks, providing objective measurements, and facilitating earlier detection of CTS. Methods: We systematically searched multiple electronic databases, including Embase, PubMed, IEEE Xplore, and Scopus, to identify relevant studies published up to January 1, 2025. Studies were included if they focused on the application of AI in US imaging for CTS diagnosis. Editorials, expert opinions, conference papers, dataset publications, and studies that did not have a clear clinical application of the AI algorithm were excluded. Results: 345 articles were identified, following abstract and full-text review by two independent reviewers, 18 manuscripts were included. Of these, thirteen studies were experimental studies, three were comparative studies, and one was a feasibility study. All eighteen studies shared the common objective of improving CTS diagnosis and/or initial assessment using AI, with shared aims ranging from median nerve segmentation (n = 12) to automated diagnosis (n = 9) and severity classification (n = 2). The majority of studies utilized deep learning approaches, particularly CNNs (n = 15), and some focused on radiomics features (n = 5) and traditional machine learning techniques. Conclusion: The integration of AI in US imaging for CTS diagnosis holds significant promise for transforming clinical practice. AI has the potential to improve diagnostic accuracy, streamline the diagnostic process, reduce variability, and ultimately lead to better patient outcomes. Further research is needed to address challenges related to dataset limitations, variability in US imaging, and ethical considerations.

Purpose Athletes face a higher risk of upper limb nerve entrapment due to repetitive stress, trauma, and biomechanics. Diagnosis is challenging, and delayed treatment can impair performance. When conservative care fails, surgery may be needed to restore function and enable return to play (RTP). Methods This systematic review adhered to PRISMA guidelines and evaluated surgical outcomes, RTP rates, and complications in athletes with upper limb nerve entrapment. A comprehensive search was conducted using MeSH terms and keywords for surgical interventions, nerve entrapment syndromes, and sports. Eligible studies included case series, cohort studies, and comparative studies that reported postoperative outcomes in athletes. Data extraction included nerve involvement, surgical techniques, clinical outcomes, and RTP rates. Results Thirty-one studies, comprising 1,297 athletes across 23 sports, were included. The most common nerve entrapments involved the ulnar nerve (50.1%), brachial plexus (39.2%), and suprascapular nerve (9.5%). Surgical interventions included ulnar nerve decompression/transposition, first rib resection with scalenectomy for thoracic outlet syndrome (TOS), and suprascapular nerve decompression. RTP rates ranged from 62 to 100%, with an average of 87%. Suprascapular nerve decompression had the highest RTP success (100%), while TOS demonstrated greater variability (62.5–97%). Functional improvements included pain reduction, increased grip strength, and enhanced patient-reported outcomes. The overall complication rate was low, but TOS procedures had the highest reoperation rates (3.8–27%). Conclusion Surgical treatment of upper limb nerve entrapment in athletes yields high RTP rates and functional recovery. Ulnar and suprascapular nerve decompressions show consistent success, while TOS surgery outcomes vary.

Purpose Subclinical peroneal neuropathy without overt foot drop has been linked to increased fall risk in adults, yet remains under reported due to subtle symptoms and lack of awareness. Patients with carpal tunnel syndrome (CTS) often experience other nerve entrapments, prompting this study to evaluate CTS (a proxy for peroneal nerve entrapment) as a significant predictor of time to first fall. Methods Data from the Merative MarketScan Research Databases (2007–2021) were used to identify adult patients using ICD-9/10 codes. Patients were stratified by CTS diagnosis and fall occurrences, with relevant comorbidities recorded. A survival analysis employing the Cox proportional hazards model assessed relationships between CTS, comorbidities, and future fall risk, accounting for changes in health status over time. Age was the time scale with CTS as a time-varying predictor. This approach isolated CTS-associated risk, while considering the natural increase in fall risk with age. Results Among 63,187,681 subjects (mean age = 52.82 years ± 7.61), 1,411,695 had a diagnosis of CTS. Of those with CTS, 45,479 patients had a future fall. Univariate analysis showed significant associations between CTS and higher rates of arthritis and diabetes, while heart disease was less prevalent. CTS increased fall risk by 25% (HR 1.25, p < .005). Heart disease was associated with a 10% increase in fall risk (HR 1.10, p < .005), while arthritis and diabetes increased fall risk by 2% (both HR 1.02, p < .005). Kaplan-Meier curve illustrated a steeper decline in survival probability for the CTS group, indicating they experienced falls at younger ages and at a higher rate than those without CTS (χ² = 4386.4, p < .001). Conclusion Prior diagnosis of CTS is associated with an increased fall risk. Providers should screen CTS patients for fall risk and implement appropriate monitoring strategies. Further investigation on the role of peroneal nerve entrapment in this increased fall risk is warranted. This study identifies a treatable cause of falls, with potential to enhance patient safety and reduce fall-related morbidity.

Postsurgical falls have significant patient and societal implications but remain challenging to identify and track. Detecting postsurgical falls is crucial to improve patient care for older adults and reduce healthcare costs. Large language models (LLMs) offer a promising solution for reliable and automated fall detection using unstructured data in clinical notes. We tested several LLM prompting approaches to postsurgical fall detection in two different healthcare systems with three open-source LLMs. The Mixtral-8×7B zero-shot had the best performance at Stanford Health Care (PPV = 0.81, recall = 0.67) and the Veterans Health Administration (PPV = 0.93, recall = 0.94). These results demonstrate that LLMs can detect falls with little to no guidance and lay groundwork for applications of LLMs in fall prediction and prevention across many different settings.

Predictive models show promise in healthcare, but their successful deployment is challenging due to limited generalizability. Current external validation often focuses on model performance with restricted feature use from the original training data, lacking insights into their suitability at external sites. Our study introduces an innovative methodology for evaluating features during both the development phase and the validation, focusing on creating and validating predictive models for post-surgery patient outcomes with improved generalizability. Electronic health records (EHRs) from 4 countries (United States, United Kingdom, Finland, and Korea) were mapped to the OMOP Common Data Model (CDM), 2008-2019. Machine learning (ML) models were developed to predict post-surgery prolonged opioid use (POU) risks using data collected 6 months before surgery. Both local and cross-site feature selection methods were applied in the development and external validation datasets. Models were developed using Observational Health Data Sciences and Informatics (OHDSI) tools and validated on separate patient cohorts. Model development included 41 929 patients, 14.6% with POU. The external validation included 31 932 (UK), 23 100 (US), 7295 (Korea), and 3934 (Finland) patients with POU of 44.2%, 22.0%, 15.8%, and 21.8%, respectively. The top-performing model, Lasso logistic regression, achieved an area under the receiver operating characteristic curve (AUROC) of 0.75 during local validation and 0.69 (SD = 0.02) (averaged) in external validation. Models trained with cross-site feature selection significantly outperformed those using only features from the development site through external validation (P < .05). Using EHRs across four countries mapped to the OMOP CDM, we developed generalizable predictive models for POU. Our approach demonstrates the significant impact of cross-site feature selection in improving model performance, underscoring the importance of incorporating diverse feature sets from various clinical settings to enhance the generalizability and utility of predictive healthcare models.